Skip to content

Reads your
any of your 57 systems.
Writes to none.

An investigator for platform and support engineering. Platform issues a read-only credential per system, 57 connectors and counting. Support asks in plain words and gets the answer with the queries that prove it. Nobody escalates to get a SELECT run.

readingGitHubCode & Reposcompare main...release/0.1.46 · 0 deploys since 22:10 · 120 msGitLabCode & Reposcompare main...release/0.1.46 · 0 deploys since 22:10 · 120 msAzure DevOpsCode & Reposcompare main...release/0.1.46 · 0 deploys since 22:10 · 120 msBitbucketCode & Reposcompare main...release/0.1.46 · 0 deploys since 22:10 · 120 msCircleCICode & Reposcompare main...release/0.1.46 · 0 deploys since 22:10 · 120 msKubernetesInfrastructureget events -n prod --field-selector involvedObject.name=edi-sync · 0 restarts · 64 msDockerInfrastructureget events -n prod --field-selector involvedObject.name=edi-sync · 0 restarts · 64 msTerraformInfrastructureget events -n prod --field-selector involvedObject.name=edi-sync · 0 restarts · 64 msConfluent / KafkaInfrastructureget events -n prod --field-selector involvedObject.name=edi-sync · 0 restarts · 64 msAWS CloudWatchCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msAWS SQS / SNSCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msAzure MonitorCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msAzureCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msGoogle Cloud LoggingCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msCloudflareCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msOktaCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msAWSCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msVercelCloudlogs filter severity>=ERROR resource=edi-sync since 02:40 · 140 msMongoDBDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msPostgreSQLDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msMySQLDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msRedisDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msClickHouseDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msSnowflakeDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msBigQueryDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msDynamoDBDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msSupabaseDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msNeonDatabasesSELECT status, count(*) FROM sync_runs WHERE started_at > now() - interval '1 hour' GROUP BY 1 · 14 msSentryObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msPrometheusObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msOpenSearchObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msElasticsearchObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msDatadogObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msGrafanaObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msPagerDutyObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msSplunkObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msNew RelicObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msDynatraceObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msHoneycombObservabilityservice=edi-sync level=error since 02:41 · 23 hits · 'EOF from partner SFTP' · 210 msAtlassianBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msStripeBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msSlackBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msNotionBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msZendeskBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msServiceNowBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msLinearBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msIntercomBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msHubSpotBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msSalesforceBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msFreshdeskBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msPostHogBusiness systemstickets.search status:open tag:edi since 02:40 · 6 tickets · 90 msNeo4jKnowledge & memorysearch 'partner SFTP EOF' · 2 notes, last edited 2026-09-30 · 40 msNeo4j knowledge graph memoryKnowledge & memorysearch 'partner SFTP EOF' · 2 notes, last edited 2026-09-30 · 40 msMemgraphKnowledge & memorysearch 'partner SFTP EOF' · 2 notes, last edited 2026-09-30 · 40 msGraphitiKnowledge & memorysearch 'partner SFTP EOF' · 2 notes, last edited 2026-09-30 · 40 msMem0Knowledge & memorysearch 'partner SFTP EOF' · 2 notes, last edited 2026-09-30 · 40 msCogneeKnowledge & memorysearch 'partner SFTP EOF' · 2 notes, last edited 2026-09-30 · 40 mswrites 0

Platform engineering
Issue read-only credentials per system, set the model key, put people on teams. Ten minutes in the portal. Every query any of them runs lands in the audit log; none of them can write.
Support engineering
Ask the console or let a ticket auto-triage. Postgres, the logs, the deploy history and the Kubernetes events get read in parallel, and the reply carries the queries. No ping to an engineer to confirm a hunch.
  • Read-only connections, credentials you issue
  • Runs on your own model key
  • Desktop app for the team, portal for admin
  • Org-wide audit log of every query
  • ₹1,760 per seat per month (≈ $20 USD), add-ons optional

One working day, as Triagic worked it

Real investigations with a clock time, in order. The clock on the left keeps up as you read.

03:12Alert rule “Failed syncs over 10”

Failed syncs (last hour) read 23 against a threshold of 10. The rule fired critical.

54 Saltgrass Apparel EDI messages rejected as duplicates, 135 orders inside. Acknowledged from the email at 03:19.

  1. mongodbsaved query of the Failed syncs (last hour) tile: rejected edi_messages23. Run read-only from the rule owner's desktop, every 15 minutes.
  2. rule23 > 10, one checkBreached. A plain comparison: no model call, nothing charged.
  3. notifycritical ladderBell and a sticky notification on the owner's desktop, email to Warehouse ops, a post in the Slack alert channel.
  4. cloudstatus before the 03:27 repeatAcknowledged by Ana at 03:19 from the email link. The repeat was not sent.

Instead of Finding out at the 07:00 shift handover, after the night's orders had missed the warehouse.

Mon 08:00Dashboards, New dashboard

“Monday floor review: on-time shipping by warehouse, orders past SLA, Reno pick rate, the Dallas freezers, inbound stuck at the dock, failed EDI syncs, invoices at risk.”

Reno on-time at 68% against 97% and up everywhere else, and 214 Reno orders past SLA.

  1. proposePropose a layout, priced $0.31 before it ranNine items, each with its source and its build price. Two unticked.
  2. postgreson-time by warehouse, last 48 h: orders, shipped_at <= sla_due_atReno 68%. Dallas, Allentown and Atlanta above 97%.
  3. mongodbpick_waves, Reno units per hour; sensor_readings, Dallas F1 and F2Reno down from about 105 to about 60 since Saturday. F2 at -16.2°C, F1 at -20.1°C.
  4. buildBuild 7 items, $0.84Seven saved read-only queries, each test-run once. From here, opening and refreshing are free.

Instead of A spreadsheet rebuilt by hand every Monday, or a request to a data team Corvane doesn't have.

Tue 09:00Report, every weekday at 09:00

The Production playbook's brief: label failures by carrier, new Sentry issues, deploys and disabled webhook endpoints since the last edition.

Every UPS label to a non-US address has failed since carrier-adapter 4.2.0 at 17:40. 389 overnight.

  1. mongodbshipments since the last edition, label_failed by carrier and destination389 UPS failures outside the US, UPS_ADDRESS_VALIDATION_FAILED, carrier code 264002.
  2. mongodbdeployments since the last editioncarrier-adapter 4.2.0 at 17:40 moved address validation to UPS's street-level API.
  3. mongodbwebhook endpoints disabled overnightBloom Botanics, after 50 consecutive 401 responses. Events are queuing.
  4. sentrynew issues and rate changes since the last editionOne new issue, matching the UPS failures. Nothing else moved.

Instead of The first ten minutes of standup spent asking whether anything broke overnight.

Tue 09:14SOC 2 readiness checkup, monthly

The auditor's evidence request: access reviews, off-boarding records, change approvals. Due Friday.

A shared Postgres superuser, and three departed users who still hold write.

  1. postgrespg_roles, role_table_grants, pg_stat_activitySix distinct client_addr values connected as postgres in the sampled hour.
  2. githubmembers, outside collaborators, deploy keys, review state of merges, 90 daysThree members with no commit in 120 days still hold write on two private repositories.
  3. ticket historyincident tickets from the last quarterThe evidence for CC7 that no other system holds.

Instead of Three weeks, 214 screenshots and evidence-final-v3.xlsx.

11:40Email from ops@arcadialabs.io

Bulk import stuck at 82% for two hours. Third time this quarter. They're “evaluating alternatives”.

Import worker OOM-killed on a 240 MB CSV. The first reply carried the cause.

  1. kubernetespod status and events, import-workerOOMKilled at 11:31.
  2. postgresthe customer's row in import_jobsRow 8412 stalled at 82%.
  3. opensearchworker logs around the stallA heap error while parsing a 240 MB file.

Instead of An apology and a promise to look into it.

14:07 UTCZendesk ticket #4821

“Checkout is failing for EU customers since ~14:00 UTC.”

The checkout-refactor deploy broke EU card charges. Support answered it. Nobody got paged.

  1. zendeskread ticket #4821Personal data is redacted before the model sees any of it.
  2. postgresSELECT status, count(*) FROM orders WHERE region='eu' GROUP BY status41 failed, 3 paid since 14:00. Ran inside a read-only transaction.
  3. sentrynew issues since 14:00312 new StripeCardError events, first seen 14:02 UTC.
  4. githubrecent releasesThe onset matches release 2026.08.02-checkout-refactor, deployed 13:58.

Instead of A message in #eng-oncall, two people, 42 minutes before anyone had looked.

Fri 16:02Asked in the Console

“How is the 16:02 bulk-export deploy doing?”

Healthy for 97% of exports. Exports over 500k rows OOM the worker.

  1. prometheusexport workers: error rate and memory, before and after 16:02Export 5xx went from 0.1% to 2.4%. worker-exports was OOMKilled six times.
  2. sentryissues first seen since the deployThe stack trace that ends up in the bug report.
  3. posthogsessions using bulk export right now, and where they drop offThree orgs affected, all on the enterprise plan.

Instead of Twenty minutes across three dashboards, or going home and hoping.

17:50GitLab code review checkup

!287 Backfill merchant analytics + retire legacy index. +412 −38, pipeline passed, approved in nine minutes.

Line 41 drops an index on a 38-million-row table, with nothing to replace it.

  1. gitlabdiff of ingest-pipeline !287, file by filemigrations/0142.sql:41 drops idx_events_merchant_created. Read through the API: no checkout, no push.
  2. gitlabapproval and pipeline historyThe author and the approver are the same account.

Instead of One review comment: “looks fine”.

Live chat · 14:09

The same outage, two minutes later, in the chat bubble on your site.

A visitor writes, the desktops of the people on duty ring, one of them accepts and investigates the chat before answering. Same reads as the ticket, same evidence, a person sends.

One snippet on your site
Paste it before the closing body tag on every page that should show the chat bubble. Allowed sites, one per line; chats from any other site are refused.
It rings who is available
Route a widget to a team, or ring every available member of the organization. Availability is a switch in the app's header.
End-to-end encrypted
Messages and files are encrypted between the visitor's browser and the agent's app. They pass through a relay on Cloudflare, and neither Triagic nor Cloudflare can read them.
Investigate this chat
Pick a playbook and the integrations to use. The same read-only reads as a ticket, every query shown, and a suggested reply that a person sends.
Save as ticket
The conversation becomes a ticket, or is added to the visitor's open one.
When nobody is available
Show a message only, or one of your live forms. Cloudflare Turnstile can check visitors are human before a chat starts.
Read the full record

Livecheckout.example

  1. A visitor is waitingRingingAccept
  2. Visitor · 14:09Our EU checkout has been failing since about 14:00 UTC. Is this on your side?
  3. 14:09 · Accepted by Maya, tier-1 support. Messages and files are encrypted between the visitor's browser and this app.
  4. Investigate this chat
  5. postgresSELECT status, count(*) FROM orders WHERE region='eu' GROUP BY status41 failed, 3 paid since 14:00. Ran inside a read-only transaction.
  6. sentrynew issues since 14:00312 new StripeCardError events, first seen 14:02 UTC.
  7. EU card charges started failing at 14:02 UTC, four minutes after the checkout-refactor rollout. US and APAC are clean.
  8. Suggested replyYou're right, and it's on us. A change we shipped at 15:58 CEST broke EU card payments. It's rolled back and payments are going through again.Maya presses Send · 14:11
  9. Save as ticket

Not on the clock

ThursdayThe PM or founder who owns the roadmapAsked in the Console14 orgs in 90 days, 3 in the last 30, and declining.
ThursdayThe founder who writes the roadmapAsked in the Console12 drafts: nine new issues and three updates to issues that already existed.
Month-endThe support lead who reports on volumeTicket trends checkup, monthlyOne cause, three months running, 74 tickets.
Any dayThree people and one shared mailboxForms + email add-onEvery request gets one ticket, one thread and one owner.

One investigator, thirteen places it works

It answers support tickets, answers questions in a console, watches your numbers and pages you when they go bad, reviews your code and your access on a schedule, and files what it finds.

Everything Triagic does
01

Intake

Where questions come in.

The Tickets queue in the Triagic desktop app. Each ticket shows its source (form, Zendesk, HubSpot, Slack, Jira), the customer, the playbook it was routed to, its status and a one-line root cause.
The Tickets queue, in the desktop app. Demo workspace.
Tickets
Zendesk, HubSpot, Jira, Slack, email threads, your own forms and manual tickets, in one queue.
Forms + emailadd-on
A hosted or embedded intake form in front, one email thread per request behind it.
Live chat
A chat bubble on your own site that rings whoever on the team is available in the app.
02

Investigate

Where the agent reads your systems and answers.

Editing the Data import playbook: routing hints, the data sources it may use (postgres, opensearch, kubernetes) and its triage instructions.
The playbook editor, in the desktop app. Demo workspace.
Auto-triage
Tickets arrive already summarized, tagged, and routed to a playbook.
Console
Ask a question in plain words and watch every query it runs.
Playbooks
One procedure per kind of ticket, each with its own data sources and teams.
Knowledge base
Your runbooks and product notes, searched during investigations.
03

Watch

What keeps checking when nobody is asking.

The Delivery operations dashboard: four KPIs with their change on the previous 30 days and a line chart of late deliveries, marked Free to view, 8 saved queries, 0 AI tokens per view, built for $0.25.
A dashboard, in the desktop app. Demo workspace.
The Alerts page in the Triagic desktop app, on the Acknowledged tab. One critical alert, Late deliveries over 8% on the Delivery operations dashboard, fired at 9.4 against a threshold of 8, repeated twice and was acknowledged by a teammate 12 minutes later. Under it, the fixed severity ladder: which severities ring the bell, send a notification, email, post to Slack and repeat until acknowledged.
The Alerts page, in the desktop app. Demo workspace.
The Reports page in the Triagic desktop app. The rail lists two series by playbook, Payments & checkout every day at 07:30 and Platform reliability every Monday at 08:00, with their editions. The open edition, 7 of 7, is today's Payments & checkout brief: card declines at baseline, refunds up 12%, 14 orders stuck in payment_pending, with the 4 queries behind it folded under How this report was made.
The Reports page, in the desktop app. Demo workspace.
Dashboards
Describe the numbers once. Opening and refreshing the dashboard never spends AI.
Alerts
Rules on dashboard numbers, written from a prompt, that notify people by severity.
Reports
A playbook on a schedule, posted as a new edition every run.
Checkupsadd-on
Scheduled read-only reviews of your code, your access and your support operation.
04

Act

How a finding leaves Triagic.

A Console thread in the desktop app. The agent called propose_issues and drafted four issues for northwind/platform. Nothing is filed until someone reviews them.
A Console thread drafting issues, in the desktop app. Demo workspace.
Issues
File a finding to GitHub or GitLab, or draft a batch of issues from one prompt.
Suggested replies
A reply drafted from the evidence. A person reads it and presses Send.

Read-only by design

The AI investigates. It doesn't get to change anything.

Postgres connections run every query inside a read-only transaction, MongoDB's read-only flag is on by default, and the Kubernetes connector explicitly disables every destructive tool. MySQL runs a server Triagic writes itself, where every query runs in a read-only transaction that is rolled back. Investigating a ticket should never mean risking the system it's about.

Every call
Listed as it runs, with its query or its arguments. If a number looks wrong, you can read how it was produced.

postgresSELECT status, count(*) FROM orders WHERE region='eu' GROUP BY status

41 failed, 3 paid since 14:00. Ran inside a read-only transaction.

From Zendesk ticket #4821, at 14:07 in the day above.
Personal data
Email addresses, phone numbers and IDs in a ticket are redacted before any of it reaches the model.
Replies
Drafted from the evidence, sent by a person. The agent has no email tool, by construction.
Code review
Reads the diff through the API. It never checks the code out and it can't push.
Alerts
Plain comparisons against a saved read-only query, checked on the rule owner's own desktop. No model, no cost per check.
01:1201:5702:4203:27failed syncs > 1003:12 fired, critical03:19 acknowledged from email
Each dot is one check: the saved query, then one comparison. From Alert rule “Failed syncs over 10”, at 03:12 in the day above.
The model
Your own OpenAI, Anthropic, Azure, Gemini, OpenRouter or Ollama key. Spend runs against your contract, not a shared account.

One plan. Every seat included.

₹1,760 per seat per month, ≈ $20 USD

14-day free trial, no credit card. Add teammates any time, up to 500 seats self-serve.

What every seat includes

  • Tickets from Zendesk, HubSpot, Jira, Slack and email
  • Live chat on your own site, switched on by request
  • Auto-triage, one ticket or a batch
  • Console, with every query shown
  • Playbooks, unlimited, drafted from a prompt
  • Knowledge base, uploaded or connected
  • Dashboards built from a prompt, free to open
  • Alerts on dashboard numbers, no cost per check
  • Reports: playbooks on a schedule
  • Issues filed to GitHub or GitLab
  • Every connector in the catalog
  • BYOK: bring your own model provider key
  • Teams, shared playbooks and credentials
  • Org-wide audit log
  • A monthly spend cap

Add-ons, per seat, only for the people who use them

  • Forms + email ₹1,760 per seat per monthDetails
  • Checkups ₹1,760 per seat per monthDetails

Approve the access once. Stop answering the same question twice.

Issue read-only credentials, connect your model key, install the desktop app for the team. Support answers from your systems with the query attached, on-call gets paged with the evidence, and you get an audit log of every read.